Abstract
Humans explore visual scenes by continually alternating rapid gaze shifts (saccades) with slow eye movements (ocular drifts). Recent work has shown that, on the retina, this oculomotor alternation yields a luminance flow with power equalized within a bandwidth that oscillates with the saccade/drift cycle (Mostofi et al, Current Biology 2020). Here we use neural modeling to investigate the possible impact of this visual input signal on retinal activity and perception. The responses of retinal ganglion cells (magno- and parvo-cellular, ON and OFF) were modeled at various eccentricities via spatiotemporal filters based on neurophysiological data. Neurons were assumed to be insensitive to purely stationary stimuli (0 Hz), hence primarily driven by luminance fluctuations elicited by eye movements. The model was exposed to luminance flows experienced by human subjects detecting a grating annulus (spatial frequency at 2 or 10 cpd; eccentricity of 0, 4 or 8°; and width of 1.5°) embedded in a natural noise field. The stimulus was displayed with variable post-saccadic exposure (50, 150, or 500 ms) contingent with the onset of a 6° saccade. Post-saccadic neural responses were cumulated by a standard decision-making stage to report the presence/absence of the grating. The model closely replicates human visual dynamics at all considered eccentricities. Responses to the luminance changes induced by eye movements cause visual sensitivity to low spatial frequency to saturate immediately after the saccade and sensitivity to high spatial frequency to continue increasing during post-saccadic drifts. These findings support the proposal that eye movements are major contributors to visual sensitivity. They indicate that quantitative models of retinal responses are sufficient for predicting the dynamics of contrast sensitivity across the visual field during natural viewing.